--- title: Who takes the ECB's money banner: Euro-area transmission map tagline: Actors, outside options, and the fair take-up of MRO and LTRO date: 2026-08-07 author: bzhmacro status: Working note flag: "€" --- # Who takes the ECB's money — *and what they would do instead* The Eurosystem now runs a demand-driven floor. It sets the deposit facility rate, offers unlimited one-week and three-month money fifteen basis points above it, and waits. Whether that design delivers a functioning money market is not a question about the size of the ECB's balance sheet. It is a question about a few dozen kinds of institution: which rate each one faces, what it would do instead, how long it needs the money for, and what collateral it owns. This note assembles those parameters and specifies the model they are for. The empirical claim is narrow and testable. Recourse to the Eurosystem responds to the spread between the market's secured rate and the MRO rate, and to almost nothing else that is currently observable. Since the corridor was narrowed in September 2024 that spread has compressed from 52 basis points to 13, and MRO outstanding has risen from under €3bn to €15bn. That is a fourfold rise in quantity for a fourfold fall in the price gap. It is also, still, one end of a demand curve. ## Where the framework stands The March 2024 operational framework review settled three things and deferred two. It settled that the deposit facility rate steers the stance; that the MRO would be priced at DFR plus 15 basis points from 18 September 2024, down from 50; and that the MRO and the three-month LTRO would remain fixed-rate full-allotment against the broad collateral list. It deferred structural longer-term operations and a structural bond portfolio to "a later stage", and it promised a review of key parameters in 2026. | Parameter | Level | Detail | | --- | ---: | --- | | Deposit facility rate | 2.25% | the steering rate | | MRO rate | 2.40% | DFR + 15bp, one week, full allotment | | 3-month LTRO | avg MRO | floating in effect, not a term rate | | Marginal lending facility | 2.65% | DFR + 40bp | | €STR | 2.19% | 6.5bp below the DFR | | Eurex GC Pooling, ECB basket | 2.29% | 3.6bp *above* the DFR | | Excess liquidity | €2,117bn | from a peak of €4,751bn on 14 November 2022 | | MRO outstanding | €21.9bn | week 31 of 2026 | | 3-month LTRO outstanding | €14.2bn | week 31 of 2026 | | Minimum reserves | €175bn | 1% of the reserve base, remunerated at 0% | Rates and quantities as at 5–6 August 2026, from the ECB Data Portal via the project's own registry. The interactive version of this table refreshes; this one does not. > [!NOTE] > One number in that table has changed character rather than level. GC Pooling used to print below the deposit facility rate essentially always — on only 3% of trading days in 2022 and 2023 was it above. In 2025 it was above on 31% of days. So far in 2026 it has been above on 88%. Secured euro cash is no longer cheaper than the central bank's floor, and the entire take-up question follows from that. ## The actors, and what each one is actually choosing The map has five columns, ordered by distance from the policy rate measured in balance sheets crossed. Only the second column can transact with the Eurosystem at all. **Banks.** The euro area's twenty banking systems collapse, for this purpose, into two behaviours. Reserve-rich dealer banks — the German, French and Dutch G-SIBs — hold reserves above their preferred level and lend them on, secured. They will never bid the MRO on price; they would bid it only if their own funding market seized. Reserve-scarce domestic banks are the marginal bidder: loan books, limited high-quality collateral, a reserve position at or below preferred. Their choice set is repo, deposits, term issuance, or the MRO, and only the last will monetise a credit claim. Two smaller bank nodes matter more than their size suggests. Euro-area branches of US, UK, Japanese and Swiss groups have deposit facility access but manage liquidity against a different home rate, which makes them the natural arbitrageur between €STR, the FX-swap implied euro rate and the DFR. And promotional banks — KfW, CDP, ICO, Bpifrance — are large, cash-rich, and heterogeneous in status: KfW is not a bank under the German Banking Act, CDP is a monetary financial institution subject to reserve requirements but outside CRR supervision, and the Spanish and French cases differ again. To the extent any of them lacks deposit facility access it lends cash in the market instead, occupying the position the Federal Home Loan Banks occupy in the United States. The status of each should be checked against its own national central bank's counterparty list before the node is used quantitatively. **Money markets.** CCP-cleared GC repo is the MRO's direct competitor and the most important number on the map. Bilateral and special repo is where scarcity in a specific bond shows up, and it is the channel through which the Eurosystem's *bond* holdings, not its lending, move money market rates. The unsecured overnight market that produces €STR is small, and prints below the DFR for a structural reason set out below. **Non-banks.** Money market funds, insurers, pension funds, hedge funds, corporate treasurers, debt agencies. None of them has a Eurosystem facility. Two partial exceptions have appeared recently and both are instructive for what they withhold: CCPs hold the great majority of their liquid resources as cash at central banks and, since the April 2025 TARGET decision, may draw overnight credit; and non-bank payment providers have been able to hold T2 and TIPS settlement balances since October 2025 — with no credit, no policy remuneration, and a cap at twice trailing peak outflow. The perimeter widened. Nobody got a new floor. **End demand.** Household and corporate deposit rates, corporate and mortgage lending rates, sovereign issuance. This is where transmission is finally measured, and where the largest wedge on the whole map sits: euro-area household overnight deposits pay 0.28% against a deposit facility rate of 2.25%. ## Why €STR prints below the floor A US money market fund with cash to place overnight places it at the Federal Reserve, through the ON RRP, at a rate a few basis points under IORB. A euro-area money fund cannot place it anywhere except a bank, a repo or a bill. The deposit facility is open to credit institutions and to nobody else. So every euro of non-bank cash that wants to earn the deposit facility rate must rent a bank balance sheet to get there, and the rent is the €STR–DFR spread. This is a design feature, not a malfunction, and the comparison makes it plain: | | Euro area | United States | | --- | ---: | ---: | | Administered floor | DFR 2.25% | IORB 3.65% | | Unsecured overnight | €STR 2.19% | EFFR 3.63% | | Spread to floor | −6.5bp | −2.0bp | | Secured overnight | GC Pooling 2.29% | SOFR 3.64% | | Non-bank facility | none | ON RRP | The euro floor is soft by construction and the dollar floor is hard by construction, and the difference is one facility. It follows that a euro-area money fund's bargaining position deteriorates whenever bank balance sheet becomes expensive — at quarter-ends, at year-end, and in any episode where the leverage ratio binds. The €STR–DFR spread is therefore a price of intermediation, and reading it as a stress indicator is a category error. ## What a bank's alternatives actually are The take-up question is a cost comparison, so it needs a complete and correctly priced option set. For a reserve-scarce euro-area bank the set is six items long. | Option | Tenor | Price | Collateral | Constraint | | --- | --- | ---: | --- | --- | | Bid the MRO | 1W | 2.40% | full Eurosystem list, incl. credit claims | uncapped; 0% LCR outflow | | Bid the 3M LTRO | 3M | avg MRO | same | term certainty at the same price | | Borrow GC repo | O/N–3M | 2.29% | ECB basket only | up to 100% LCR outflow on non-HQLA | | Pay up for term deposits | 1–12M | 2.19% | none | reprices the back book | | Issue a covered bond | 3–10y | swaps +25bp | encumbers the cover pool | competes with the MRO for the same assets | | Issue senior non-preferred | 5–10y | swaps +95bp | none | MREL makes a minimum volume compulsory | Deposit and repo rates are measured; the two issuance spreads are assumptions, flagged as such wherever they appear, and are among the parameters a proper model would have to identify rather than assume. Three features of this table are easy to miss. First, the MRO is *not* generally the cheapest option and is not meant to be. At 2.40% against GC at 2.29% it is eleven basis points expensive. That is the framework working: the Governing Council chose fifteen basis points precisely to make recourse occasional rather than either continuous or unthinkable. Second, the covered bond and the MRO are substitutes for the same collateral, not complements. A bank has one cover pool. Pledging it to the Eurosystem and encumbering it in a covered bond are alternative uses of a single asset, which means the covered bond spread belongs in the MRO decision even though it is a five-year instrument and the MRO is a one-week one. Third, only the MRO transforms a credit claim into central bank money. No repo counterparty will take a loan book. This is the asset-transformation motive, and it is the reason take-up is not a pure function of the price gap — a bank with no marketable collateral will bid at spreads where a bank with Bunds will not. ECB Working Paper 2439, on TLTRO participation, is the empirical anchor here: take-up scaled with the ability to pledge illiquid assets, not only with funding need. ## The collateral question The eligible pool is narrowing at exactly the moment reserves are. Three changes point the same way. Additional credit claim frameworks, introduced in 2011 and adopted by seventeen national central banks, are being phased out; the Governing Council took a further step in June 2026, with technical implementation running into 2027. The revised risk-control framework, in force from November 2026, gives own-used covered bonds a dedicated haircut schedule rather than an add-on, and gives retained ABS tighter dedicated treatment under a broader definition of "retained". Against those, marketable assets issued through DLT-based CSDs became eligible on 30 March 2026 — a genuine widening, and a small one. The net effect is that the collateral a bank can bid the MRO against is shrinking while its need to bid is growing. Whether that matters depends entirely on the distribution of collateral inventories across banks, which is not published. The opposite mechanism deserves equal weight. APP and PEPP run-off destroys reserves and returns bonds to the float simultaneously. Scarcity and ampleness therefore move in opposite directions, which is why the 2022–23 episode — when German collateral traded far below the deposit facility rate — unwound as quantitative tightening progressed rather than in spite of it. Any model that treats the balance sheet as a single tightening variable will get the repo market wrong. ## What the literature actually establishes Four strands, of unequal strength. **Price.** Uncontroversial and mechanical: recourse falls as the outside option cheapens. This is the logic of the March 2024 narrowing itself, and the IMF and Bruegel assessments of the framework review both treat it as the binding consideration. **Collateral opportunity cost.** Well established. Pledging a bond forgoes the repo rate and any specialness it commands. The ESRB and ECB work on collateral scarcity premia establishes the mechanism from the other direction: central bank *purchases*, by removing specific collateral from the float, push specific-collateral repo away from the policy rate. **Regulatory treatment.** Concrete, quantifiable, and usually left out of the comparison. The asymmetry is in the LCR and not, as is often asserted, in the NSFR. Secured funding from the domestic central bank maturing within thirty days carries a 0% outflow rate whatever the collateral; market repo of the same maturity against collateral that is not HQLA carries up to 100%, with a 25% cap only for a domestic sovereign, qualifying public-sector or multilateral counterparty. For a bank funding illiquid assets at short tenor that wedge is worth considerably more than fifteen basis points, and it applies to the MRO's own one-week tenor. The NSFR, by contrast, does not discriminate. Its encumbrance relief — an asset encumbered for less than six months keeps the required stable funding factor of the equivalent unencumbered asset — turns on the remaining term of encumbrance, not on who the counterparty is, so it treats a pledge to the Eurosystem and a short market repo identically. And on the funding side the NSFR explicitly brackets central banks with financial institutions: both attract 0% available stable funding inside six months. Any claim that the NSFR tilts a bank toward the MRO is wrong, and it is a common enough error to be worth stating plainly. **Stigma.** The weakest link. The euro-area evidence is qualitative: the ECB has never described its operations as a backstop, has said explicitly that their use is "an integral part" of policy implementation, and the marginal lending facility has been drawn on without the reticence attached to the Fed's discount window. There is no euro-area equivalent of the Furfine or Armantier–Ghysels–Sarkar–Shrader identification of a stigma premium. > [!IMPORTANT] > Stigma is endogenous to scarcity. Recourse is uninformative when everyone is doing it and informative when nobody is. The ECB's December 2025 decision to have counterparties test their access to the MRO and the 3-month LTRO at least annually is best read as pre-empting exactly this — and it happens to create the identification the literature lacks, by separating operational recourse from distressed recourse in the data for the first time. ## The take-up curve Put the price on the horizontal axis and the quantity on the vertical, one point per month, and the relationship the framework depends on becomes visible. ![Monthly average of Eurex GC Pooling minus the MRO rate, against MRO outstanding, January 2019 to July 2026. The 3-month LTRO is excluded because that series carried TLTRO-III until December 2024.](takeup-curve.png) Two clusters, and the gap between them is the September 2024 corridor narrowing. Before it, the market rate sat an average 52 basis points below the MRO and MRO outstanding averaged €2.6bn — a rounding error against a balance sheet then above €8trn. Since it, the gap has averaged 16 basis points and MRO outstanding €10.6bn. July 2026 sits at 13 basis points and €15.3bn, the narrowest gap and the largest recourse in the sample. The slope is the right sign and the magnitude is plausible. It is also entirely identified off one policy change, which means the elasticity is not separable from whatever else the narrowing did. And the vertical line at which the market rate equals the MRO rate has never been observed: it is off the chart, to the right. Every statement about what happens when the gap closes is therefore an extrapolation. ## Towards a fair take-up State the decision properly. A bank funds a euro of assets over a horizon by the cheapest available route, subject to its collateral and its binding constraint. The all-in cost of the MRO route is **c₍MRO₎ = DFR + 15bp + κ·(repo rate forgone on the pledged asset) + h·(overcollateralisation cost) + σ·(stigma) − λ·(LCR relief versus market repo)** Only the first term is observed directly. The map supplies κ from the collateral-specific MMSR repo series, h from the published haircut schedule, and leaves σ and λ as the two free parameters. Fair take-up is then the mass of funding need whose next-best alternative costs more than c₍MRO₎ — which requires a distribution of collateral and funding need across banks, not an aggregate. Four gaps stand between the specification and the estimate, and three of them are deliberate on the ECB's part. - **Collateral inventories by counterparty.** The Eurosystem publishes collateral put forward by asset class in aggregate. Without a distribution across banks the asset-transformation motive cannot be sized, and it is precisely the motive that decouples take-up from the price gap. - **Take-up by counterparty.** Operation results give allotted amounts and bidder counts, never identities or concentration. This is the single largest gap. A €15bn MRO taken by four hundred banks and a €15bn MRO taken by six banks are different observations about the framework, and they are indistinguishable in the published data. - **Term secured pricing.** The 3-month LTRO's real competitors are 3-month repo and 3-month CD issuance. The published secured benchmarks are overnight. Term has to be built from OIS plus an assumed secured basis, and every such assumption is an input the model should identify rather than impose. - **An identified stigma parameter.** See above. The annual access-testing regime is the natural experiment; it needs a few more years of data. Until those close, the honest position is that the price elasticity of MRO take-up can be estimated from twenty-three monthly observations since the corridor change, and that a structural model of who bids and why cannot. The first is worth doing and is what the interactive page now supports. The second requires data the Eurosystem has and does not publish. ## Three models of the same market The map is a picture of constraints. Turning it into arithmetic needs one idea: an agent's **reservation rate** is the best it can do without you, and every rate in this system is a split of the gap between two reservation rates, net of what it costs whoever stands in the middle. Three models formalise that at increasing cost. **Model A — bargaining**, after Bech and Klee (2011), who asked why the effective federal funds rate traded below interest on reserves and answered that the agencies lending in that market could not earn it. A non-bank lending overnight to a bank receives **rᵢ = rᵢᵒᵘᵗ + θᵢ·(DFR − c − rᵢᵒᵘᵗ)** where c is what the balance sheet costs the bank. The result worth keeping is what happens at θᵢ = 1: even a money fund with complete bargaining power receives only DFR − c. **The euro floor has an irreducible leak equal to the cost of intermediation, and no amount of negotiation closes it.** The only policy that does is one that removes the intermediary. The same c bounds the secured market from the other side, because a bank lends cash in repo only if it earns at least DFR + c. That gives two markets and two parameters with no circularity: c is identified from the GC premium over the deposit facility, and bargaining power is then identified from €STR. It also gives a restriction the model can fail. It fails. Both benchmarks should sit inside DFR ± c, a band 4.5 basis points wide on the current calibration; they are 10.1 basis points apart. Balance-sheet cost explains under half the distance between the euro area's two overnight benchmarks. **Model B — directed search**, after Armenter and Lester (2017), supplies the other half. Banks post rates, lenders choose which queue to join, and how much of the surplus a lender keeps is an outcome rather than a primitive. Fitting it to the same €STR moment puts the representative non-bank's outside option at **DFR − 11bp**, which is an estimate of the thing the euro area does not publish and the United States does not need to, because there the ON RRP sets it by decree. Its sharpest result is a proposition rather than a number. Trading requires a strictly positive spread between the deposit facility rate and the non-banks' outside option; at zero spread no bank finds it worth entering and the market does not shrink but *ceases*. So a euro-area ON RRP is not a free improvement to the floor. At DFR − 4bp it would hand non-bank lenders about five basis points and leave the intermediated market broadly intact; at the DFR itself it would close the unsecured market entirely. That trade-off is on the page as a table, and it is the honest form of a question usually asked as though only one side existed. **Model C — OTC search**, after Afonso and Lagos (2015), adds time. Every simulation starts with exactly enough reserves in aggregate and still ends with banks at the penalty rate, because they could not find each other fast enough. That is friction F2 priced: at a plausible meeting rate, roughly nine per cent of banks close short and pay an average of 1.3 basis points for it. It is also the honest explanation of why spreads widen at quarter-ends — not a signal about the stance, but this model with the meeting rate falling as dealer balance sheet becomes expensive. The three are consistent where they should be. Model C's balanced-market rate and Model A's interbank leg are the same object written twice, and the test suite checks they agree to machine precision. ## What the take-up curve is actually worth Fitting **T(s) = a + b·e^(s/v)** to the ninety-one monthly observations gives a structural floor of €1.9bn, and v = 13 basis points — every thirteen basis points of compression multiplies discretionary recourse by e. The elasticity has a one-line form, dT/ds = (T − a)/v, so at the current gap one more basis point of compression buys about **€0.8bn** of additional take-up. Two things about that fit matter more than the point estimates. It passes a real test. The sample straddles September 2024, when the MRO−DFR spread was cut from fifty basis points to fifteen. The model says only the gap matters — not the corridor width, not the level of excess liquidity. A Chow test across that break gives F = 2.33 against a five per cent critical value of 2.71: a single demand curve in the price gap spans a policy regime change. And it cannot answer the question everyone wants answered. The gap has never been narrower than about thirteen basis points, so where take-up goes when it closes is an exponential extrapolated across an interval as wide as the one it was fitted on. The curve says roughly €30bn at a zero gap. The defensible statement is narrower: the slope near the current gap is well measured, the level at closure is not, and the difference between those two claims is exactly the difference between a model and a forecast. ## Fragmentation: the collateral is not fungible Everything above treats euro collateral as one pool. It is not, for two reasons that compound. **Banks hold their own government.** From the final EBA Transparency Exercise, the share of each banking system's sovereign book issued by its own government: Greece 53%, Spain 50%, Italy 50%, Finland 49%, France 47%, Belgium 43%, Netherlands 40%, Germany 37%. Those numbers cannot be compared with each other as they stand, and the first version of this note compared them against the wrong thing — one twentieth, on the reasoning that there are twenty euro-area issuers. That benchmark is meaningless. No bank was ever going to hold as much Estonian paper as German paper, because Estonia has issued about €5bn of government debt securities and Germany about €2.3 trillion. A neutral portfolio is not equal-weighted; it holds each issuer in proportion to what that issuer has actually put in the market. So the benchmark used here is each sovereign's share of euro-area general government debt securities outstanding — Eurostat's quarterly government debt, instrument F3, €12.0trn across the twenty-one member states at 2026-Q1. France is 26% of it, Italy 22%, Germany 19%, Spain 13%; Greece is 0.8% and Estonia 0.05%. Home bias then becomes an **overweight**: what a banking system holds of its own government, as a share of its euro-area book, divided by what a proportional holder would hold. | | own government, share of the euro-area book | neutral weight | overweight | | --- | ---: | ---: | ---: | | Greece | 57% | 0.81% | 71× | | Finland | 74% | 1.63% | 45× | | Netherlands | 52% | 3.67% | 14× | | Belgium | 65% | 5.12% | 13× | | Ireland | 14% | 1.23% | 11× | | **Spain** | **72%** | **12.99%** | **5.5×** | | **Germany** | **54%** | **19.17%** | **2.8×** | | **Italy** | **61%** | **22.01%** | **2.8×** | | **France** | **71%** | **26.42%** | **2.7×** | Scaling reverses part of the story and sharpens the rest. On the raw share, German banks looked *less* home-biased than Italian ones — 37% against 50% — and that reading was mostly an artefact of Germany issuing a fifth of everything in the market. Measured against issuance, Germany and Italy are indistinguishable at 2.8×, France is marginally the most proportional of the four large systems, and Spain at 5.5× is the genuine outlier. What survives unchanged is that the large systems are moderately overweight and the small ones are wildly so: a Greek bank holds its own government at seventy times a neutral weight, which is a different phenomenon from a French bank holding France at under three. That matters for collateral because concentration means the pool re-prices together. A banking system whose euro-area sovereign book is half one issuer has a haircut that moves as one when that issuer's rating steps — and the banks affected are precisely those whose other collateral is also domestic. The scaled measure adds that this is not a periphery problem: 2.8× on a €432bn Italian book and 2.8× on a €350bn German one are the same behaviour, and only one of them sits next to a haircut boundary. **What the sample does and does not cover.** The EBA population is the 119 largest EU banks, consolidated. Germany's savings banks and cooperatives — several hundred institutions, essentially domestic — are not in it, which raises the obvious worry that the German figure is a selection artefact of a three-pillar system. It was checked against ECB balance-sheet statistics, which cover every resident monetary financial institution rather than a sample. On holdings of debt *securities* the two measures agree closely for Germany (53.4% against the EBA's 53.8% on a euro-area denominator), so the missing institutions do not explain the German number. Where the all-bank measure diverges is once loans to government are included, which is where the savings banks are concentrated — and that gap is larger in France and Italy than in Germany. Loans are not collateral, so for this note the securities measure is the relevant one, but the caveat belongs on the page. **The Eurosystem prices the paper differently.** Eligibility requires credit quality step 3 — BBB−, Baa3, BBB (low) — and Article 84 applies the **first-best rule**: where several accepted agencies rate an asset, the best rating governs. Since Scope Ratings was accepted in November 2023 there are five accepted agencies, so a sovereign becomes ineligible only when all five have abandoned it. This is routinely got wrong in the other direction. Italy in October 2018, when Moody's cut to Baa3, was widely described as one notch from losing ECB eligibility. Under the first-best rule it was three, because DBRS had it at BBB (high). The test suite uses that episode as a fixture. ## The cliff that actually binds The threshold everybody watches is far away. The one that binds is the step from CQS2 to CQS3 — A− to BBB+, a single notch — where Category I haircuts roughly quintuple. A five-to-seven year government bond goes from a 2.0% haircut to 10.0%. And the first-best rule offers no protection here at all. It guarantees that *some* agency keeps you eligible. It does not guarantee that *any* agency keeps you in a given step. So a sovereign's entire haircut treatment can rest on one agency, with no second rating to take over when that one moves. Italy is the clean case, and the numbers are large: | | today | one notch | | --- | ---: | ---: | | Governing rating | A (low), DBRS | BBB (high) | | Credit quality step | CQS2 | CQS3 | | Weighted haircut on bank-held Italian paper | 2.4% | 9.6% | | Collateral value, on €483bn held | €471bn | €437bn | | **Collateral value destroyed** | | **€34bn** | S&P and Fitch have Italy at BBB+, Moody's at Baa2 and Scope at BBB+ — four of the five already in step 3. DBRS alone, at A (low) since 17 October 2025, holds the whole stock of Italian government paper in the cheaper bracket. One DBRS notch destroys about €34bn of collateral value without any other agency moving, without any change in Italy's fundamentals as three of five agencies already assess them, and without a single euro becoming ineligible. Estonia's CQS1 rests on DBRS alone, which costs almost nothing because the CQS1→CQS2 step is small. Everywhere else at least two agencies sit at the governing step, and one downgrade changes nothing. That asymmetry — a hard threshold that is far away and heavily protected, a soft step that is close and not protected at all — is the part of euro-area collateral risk that aggregate analysis misses. ### It has happened, and to the same country The obvious objection to all of this is that it is hypothetical. It is not. Replaying every rating action in the ESMA register through the first-best rule gives the exact dates on which a euro-area sovereign's governing credit quality step changed, and there are ten of them in eleven years: | Date | Country | Move | Caused by | | --- | --- | --- | --- | | 13 Jan 2017 | **Italy** | **CQS2 → CQS3** | DBRS cut to BBB (high) | | 14 Sep 2018 | Cyprus | CQS4 → CQS3, regains eligibility | S&P upgrade to BBB− | | 22 Mar 2019 | Croatia | CQS4 → CQS3 | S&P upgrade to BBB− | | 14 Jun 2019 | Slovenia | CQS2 → CQS1 | S&P upgrade to AA− | | 29 Nov 2019 | Ireland | CQS2 → CQS1 | S&P upgrade to AA− | | 26 Aug 2022 | Portugal | CQS3 → CQS2 | DBRS upgrade to A (low) | | 8 Sep 2023 | Greece | CQS4 → CQS3, regains eligibility | DBRS upgrade to BBB (low) | | 13 Sep 2024 | Croatia | CQS3 → CQS2 | S&P upgrade to A− | | 25 Oct 2024 | Cyprus | CQS3 → CQS2 | Scope upgrade to A− | | 17 Oct 2025 | **Italy** | **CQS3 → CQS2** | DBRS upgrade to A (low) | Three things fall out of that table. **Every crossing was caused by a single agency acting alone**, and in nine of the ten that agency was the only one at the resulting step — the first-best rule protects against the worst rater and offers nothing when the best rater moves. **DBRS caused four of the ten**, including both Italian ones. And the January 2017 event is the precedent: it is a CQS2→CQS3 crossing, in the modern graduated haircut regime, on the country that is in the same position today. The set-up in 2017 was the one Italy is in now, in reverse. Fitch cut Italy to BBB+ on 8 March 2013, which left DBRS's A (low) as the only rating holding Italy in CQS2. That configuration lasted three years and ten months. On 13 January 2017 DBRS downgraded Italy to BBB (high), the governing step moved to CQS3 the same day, and the haircut tier applied to every BTP, BOT and CCT changed with it — reported at the time as 2% to 10% on a five-year BTP and 0.5% to 6% on bills, consistent with the ECB's own published range for that month. No accepted agency was at or above A−; nothing absorbed it. DBRS took rating actions on eighteen Italian structured-finance transactions the same day and cut Intesa Sanpaolo's senior rating, so the repricing was not confined to the sovereign line. On 17 October 2025 the same agency upgraded Italy back to A (low), and Italy is once again in CQS2 on DBRS alone. The configuration that broke in 2017 has been reassembled. Portugal is the other case worth knowing, because it shows the same structure one tier lower. DBRS crossed Portugal from A (low) to BBB (high) on 24 May 2011 — five months after the graduated CQS3 schedule came into force, and the earliest comparable precedent. From January 2012 the other three accepted agencies had Portugal below investment grade, so DBRS's BBB (low) was the only thing keeping Portuguese government bonds eligible at all, and its scheduled review dates became macro events: contemporaneous reporting treated 21 October 2016 as a binary for eligibility and for PSPP inclusion, and the Portuguese finance ministry was publicly briefing on it a fortnight beforehand. Then in August 2022 DBRS alone took Portugal back to A (low) and cut its haircuts by roughly four fifths. Two caveats on the precedent, and they cut in opposite directions. There is no published event study of 13 January 2017 that isolates the haircut change from everything else happening to Italian spreads that winter, so the price impact is not measured; the closest quantification is structural, from Cassola and Koulischer, whose model implies a five-percentage-point haircut increase on low-rated collateral reduces its use by about a tenth and raises the funding-cost spread by about five per cent. And in the other direction, the ECB has shown it will suspend the *eligibility* cliff when it wants to: on 22 April 2020 it grandfathered collateral down to CQS5. But it was explicit in the same act that assets falling below the minimum "will be subject to haircuts based on their actual ratings". The eligibility threshold has a demonstrated backstop. The haircut step has never been waived for a sovereign outside a programme. ## What this does to repo, and how confident to be A downgrade does not change who owns the bond; it changes how much cash the bond raises. Treating the lost collateral value as a contraction in usable float, and mapping float to specialness with a constant elasticity, the Italian one-notch case is worth roughly nine basis points of specialness against general collateral at the default elasticity. That last number is a model and the ones before it are arithmetic, and the distinction is worth keeping. No published euro-area estimate of the specialness elasticity was found, so it is a prior anchored on the 2022 collateral squeeze rather than a fitted parameter, and it is a lever on the page. Set it to zero and the euro amounts remain. The same machinery answers the rebalancing question, and reframing the benchmark changes the answer. The question is not "what if everyone held less of their own government" — that was the common-home-bias target, and it was arbitrary. It is "what if everyone held the market portfolio", each issuer at its share of debt securities outstanding. Full convergence would require euro-area banks to **buy** about €175bn of German paper and €89bn of Italian, and to **sell** about €108bn of Spanish, €58bn of French and €51bn of Belgian. Measured against issuance the banking system is not uniformly long its own government; it is short the largest issuer. At the default elasticity the Spanish sale is worth roughly −29bp of specialness and the Italian purchase +22bp, on float changes of +24% and −19%. The German number is larger than any elasticity should be asked to price: banks in the EBA sample hold €331bn of Bunds against €2.3trn outstanding, so a €175bn purchase is a 53% change in the float proxy and the row is flagged rather than reported. Ireland, Luxembourg and Lithuania are flagged for the same reason. The page carries a convergence lever so a quarter or a half of the move can be read instead, which stays inside the linear range for every large issuer. Four caveats travel with all of this. The EBA discontinued the Transparency Exercise after June 2025, so the holdings matrix is a final vintage and ages from here. Holdings are gross carrying amounts, so a hedged position looks identical to an unhedged one. Consolidation puts a group's foreign subsidiaries in its home country's row, flattering diversification. And the ratings now come from the regulatory register rather than from press releases: an audit against ESMA corrected eleven of a hundred hand-assembled entries and filled thirty-four gaps, with ten of the eleven errors sitting in entries already flagged as aggregator-sourced. ## Why this is the ECB's own trigger variable The last point is the reason the exercise is not academic. Schnabel's November 2025 framing sets an explicit sequence: structural longer-term operations begin only after "a persistent and broad-based rise in the take-up of standard refinancing operations", and a structural bond portfolio only after that. The size of the future Eurosystem balance sheet is therefore conditioned on the behaviour this map describes, and ECB staff's own range for steady-state reserve demand — somewhere between €600bn and €2.2trn — is wide enough to make clear that the ECB does not know the answer either. MRO outstanding at €21.9bn against a €5.9trn balance sheet is not yet that rise. But the direction is unambiguous: €5.5bn two years ago, €10.2bn one year ago, €21.9bn now, against a price gap that has fallen from 52 basis points to 13. On the current path the trigger arrives before the collateral pool finishes narrowing, and those two dates are the ones worth watching. ## Sources Primary, in the order they bear on the argument. - **Changes to the operational framework for implementing monetary policy** — ECB Governing Council, 13 March 2024. - **The Eurosystem's operational framework** — Isabel Schnabel, 14 March 2024. - **Towards a new Eurosystem balance sheet** — Isabel Schnabel, ECB Conference on Money Markets, 6 November 2025. Sequencing, the €600bn–€2.2trn range, the case for a short-duration structural portfolio. - **Managing liquidity in a changing environment** — Claudia Buch and Isabel Schnabel, ECB Blog, 18 March 2025. The supervisory side of operational readiness. - **How banks are adjusting to declining reserves** — ECB Blog, 2 April 2026. Redistribution through term repo; the rising share of repo trading above the DFR. - **The Eurosystem collateral framework explained** — Bindseil, Corsi, Sahel and Visser, ECB Occasional Paper 189, 2017. - **Who takes the ECB's targeted funding?** — ECB Working Paper 2439. The asset-transformation motive. - **Collateral scarcity premia in euro area repo markets** — ESRB Working Paper 55. - **Is stigma attached to the European Central Bank's marginal lending facility?** — Liberty Street Economics, Federal Reserve Bank of New York, April 2018. - **The mechanics of a graceful exit: interest on reserves and segmentation in the federal funds market** — Bech and Klee, Journal of Monetary Economics 58(5), 2011 (NY Fed Staff Report 416). Model A. - **Excess reserves and monetary policy implementation** — Armenter and Lester, Review of Economic Dynamics 23, 2017. Model B; the shutdown proposition and the facility experiment. - **Trade dynamics in the market for federal funds** — Afonso and Lagos, Econometrica 83(1), 2015. Model C; the even-split bargaining result is their Proposition 2. - **European Rating Platform** — ESMA. The regulatory record of every rating action reported by EU-registered credit rating agencies under Regulation (EC) 1060/2009, from 1 July 2015. Source for every sovereign rating and date used here. - **EU-wide Transparency Exercise 2025** — European Banking Authority. Bank-by-bank sovereign exposures by counterparty country; `tr_sov.csv`, item 2520810, June 2025 reference date, 119 banks. Discontinued after this vintage in favour of the Pillar 3 Data Hub. - **Eurosystem eligible assets database** — ECB, published each TARGET business day. Per-ISIN eligibility, haircut category and applied haircut. - **Guideline (EU) 2015/510 (ECB/2014/60), Articles 82 and 84** — the CQS3 threshold and the first-best rule. Article 82 also carries the second-best rule and CQS2 floor for ABS. - **Guideline (EU) 2016/65 (ECB/2015/35), Annex** — the valuation haircut schedule in force since 29 June 2023, revised from November 2026 at the earliest. - **Quarterly government debt (`gov_10q_ggdebt`)** — Eurostat. General government (S13) debt securities (F3), face value, consolidated. The neutral benchmark for home bias: each sovereign's share of what is actually outstanding. - **Balance sheet items (BSI)** — ECB Data Portal. Holdings of general government debt securities by *all* resident monetary financial institutions, used to test whether the EBA sample's exclusion of Germany's savings and cooperative banks distorts the German home-bias figure. - **The collateral framework of the Eurosystem: 20 years of experience** — ECB Occasional Paper 312, March 2023. Annex II dates the haircut schedules, including the introduction of a graduated CQS3 schedule on 1 January 2011 in place of the flat 5% add-on. - **Collateral haircuts and bank lending** — Cassola and Koulischer, Journal of Financial Stability, 2019. The closest thing to a quantification of what a haircut step does to the use of the collateral and to funding costs. - **ECB takes steps to mitigate impact of possible rating downgrades on collateral availability** — ECB, 22 April 2020. Eligibility grandfathered to CQS5, and the explicit statement that haircuts still follow the actual rating. - **Acceptance of Scope Ratings in the ECAF** — ECB, 10 November 2023. The fifth accepted agency, and one more rating that has to fail before a sovereign becomes ineligible. - **Basel III: the Liquidity Coverage Ratio and liquidity risk monitoring tools** — BCBS 238, January 2013, paragraphs 113–115 and the table thereunder. The 0% outflow rate on secured funding from the domestic central bank, against up to 100% on non-HQLA market repo, is friction F5. - **Commission Delegated Regulation (EU) 2015/61, Article 28(3)–(4)** — the EU codification of the same asymmetry, more explicit than Basel: 0% for the domestic central bank, 25% for a qualifying sovereign or public-sector counterparty, up to 100% otherwise. - **Basel III: the Net Stable Funding Ratio** — BCBS d295, October 2014, paragraphs 24(d), 25(a) and 31. Cited for what it does *not* say: no counterparty-based advantage for central bank funding. - **Euro Money Market Study 2024** — ECB, April 2025. Segment sizes and turnover. - **Remuneration of government deposits** — ECB, 17 April 2024. The €STR−20bp ceiling behind friction F9. - **Decision (EU) 2025/222** — non-bank payment service provider access to Eurosystem payment systems and central bank accounts.